Integrative and personalized QSAR analysis in cancer by kernelized Bayesian matrix factorization.

Area of Science:

• Computational biology
• Cheminformatics
• Pharmacogenomics

Background:

• Quantitative structure-activity relationship (QSAR) models traditionally predict drug activity for single cell lines.
• Large-scale drug sensitivity data now enables more comprehensive predictive modeling.
• Existing QSAR methods have limitations in predicting responses across diverse cell lines and for novel drugs.

Purpose of the Study:

• To develop an integrative and personalized QSAR approach for predicting anticancer drug responses.
• To simultaneously predict drug responses for multiple known cancer cell lines.
• To predict drug responses for new cancer cell lines using novel drugs.

Main Methods:

• Application of a novel kernelized Bayesian matrix factorization method.
• Integration of chemical drug descriptors, genomic cell line features, and drug target information.
• Development of both integrative QSAR for multiple cell lines and personalized QSAR for new cell lines.

Main Results:

• Demonstrated usefulness in predicting drug responses across 116 anticancer drugs and 650 cell lines.
• Analyzed the importance of various drug features for response prediction.
• Generated a comprehensive drug response map to assess treatment potential and range.

Conclusions:

• The novel method effectively predicts anticancer drug responses in integrative and personalized scenarios.
• Genomic and target information significantly improve predictive performance.
• The developed drug response map aids in identifying promising anticancer drugs and their therapeutic applications.